network model tunnel ventilation

Multiscale Modelling of Tunnel Ventilation Flows and Fires

FRANCESCO COLELLA

Thesis submitted for the degree of Doctor of Philosophy

Politecnico di Torino, Dipartimento di Energetica.

May 2010

Francesco Colella, 2010

III

DECLARATION This thesis and the research described and reported herein have been completed solely by Francesco Colella under the supervision of Professor Romano Borchiellini, Dr Vittorio Verda, Dr Guillermo Rein and Professor Jose L. Torero. Where other sources are quoted, full references are given.

Francesco Colella

TABLE OF CONTENTS

1 INTRODUCTION 1

1.1. Introduction 1

1.2. Fundamentals of tunnel fires 2

1.3. The role of the ventilation system 7 1.3.1. Natural ventilation systems 8

1.3.2. Mechanical ventilation systems 9 1.3.3. Hybrid ventilation systems 13

1.4. Interaction between fire and ventilation system 14 1.4.1. Ventilation velocity and back-layering 14

1.5. Analysis of tunnel ventilation systems and fires 22 1.5.1. Small and large scale experiments 23 1.5.2. Numerical modelling 23

1.6. Test cases 26 1.6.1. Case A: Frejus Tunnel, Bardonecchia (It) 26 1.6.2. Case B: Norfolk road Tunnels, Sydney (Au) 26 1.6.3. Case C: Wu-Bakar small scale tunnel 27 1.6.4. Case D: Dartford Tunnels, London (UK) 27 1.6.5. Case E: Test case tunnel 27

2 ONE-DIMENSIONAL MODELLING 29

2.1. Introduction 29

2.2. Literature overview 30

2.3. Typical mathematical formulation for 1D models 32 2.3.1. Topological representation 32

2.3.2. Fluid dynamics model 33

2.3.3. Thermal model 35 2.3.4. Steady state problem 36 2.3.5. Time dependent problem 40 2.3.6. Solving algorithm 42

VI

2.3.7. Typical input parameters and boundary conditions 44

2.4. A case study: the Frejus Tunnel 47

2.5. Concluding remarks 51

3 CFD MODELLING 53


3.2. Literature overview 55

3.3. Governing equations 67

3.4. Turbulence modelling 68

3.5. Boundary conditions 74 3.5.1. Pressure boundary conditions 74 3.5.2. Velocity boundary conditions 75 3.5.3. Wall boundary conditions 75 3.5.4. Boundary conditions for the transport equations of turbulent quantities 78 3.5.5. Fire representation 79 3.5.6. Jet Fan representation 82

3.6. Numerical features 83

3.7. Case Studies 85 3.7.1. Ventilation flows in the Norfolk road Tunnels 85 3.7.2. Assessment of the mesh requirements 87

3.7.3. Simulations of the ventilation scenarios and comparison to on-site measurements 89 3.7.4. Critical velocity calculation 93 3.7.5. Assessment of the mesh requirements 94 3.7.6. Critical velocity results 97 3.7.7. Effect of the fire Froude number on the critical velocity 102

3.8. Concluding Remarks 103

4 FUNDAMENTALS OF MULTISCALE COMPUTING 105


4.2. Fundamental of domain decomposition methods 108

VII

4.3. Formulation of the multiscale problem 112

4.4. Coupling technique 114 4.4.1. Direct coupling 114

4.4.2. Indirect coupling 119


5 MULTISCALE MODELLING OF TUNNEL VENTILATION FLOWS 121


5.2. A case study: the Dartford tunnels 122

5.3. Overview on the experimental setups 125

5.4. Characterization of the jet fan discharge cone 126 5.4.1. Assessment of the mesh requirements 127 5.4.2. Effect of the 1D-CFD interface location 130 5.4.3. Comparison to experimental data 132

5.5. Characterization of the ventilation system 135 5.5.1. Calculation of the jet fan characteristic curves 135 5.5.2. Comparison to experimental data 138 5.5.3. Analysis of all the ventilation strategies 139 5.5.4. Assessment of the redundancy in the Dartford Tunnels 140


6 MULTISCALE MODELLING OF TUNNEL FIRES 145


6.2. A case study: a modern tunnel 1.2 km in length 146

6.3. Characterization of the fire near field 147 6.3.1. Assessment of the mesh requirements 148 6.3.2. Effect of the 1D-CFD interface location 151 6.3.3. Comparison to full CFD solutions 155

6.4. Characterization of the ventilation system performance 161 6.4.1. Calculation of the fan and fire characteristic curves 161

VIII

6.4.2. Comparison to full CFD solutions 162 6.4.3. A note of the fire throttling effect 165


7 MULTISCALE MODELLING OF TIME-DEPENDENT TUNNEL VENTILATION FLOWS AND FIRES 169


7.2. A case study: a modern tunnel 1.2 km in length 170

7.3. Multiscale model results 175


8 CONCLUSIONS AND FUTURE WORKS 185

IX

LIST OF FIGURES

Figure 1: Typical traffic flow and induced ventilation in the 1.8 tunnel in Taipei City. Traffic density and induced ventilation as presented in [23] 8

Figure 2: A schematic of a Saccardo longitudinal ventilation system [26] 10

Figure 3: A schematic of a jet fan longitudinal ventilation system [26] 10

Figure 4: A schematic of a fully transverse ventilation system 12

Figure 5: A schematic of a supply semi-transverse ventilation system 13

Figure 6: A schematic of a exhaust semi-transverse ventilation system 13

Figure 7: Photograph of a small scale tunnel fire during the occurrence of back-layering. The fire size in

15 kW. The tunnel has an arched cross section (width 274mm, height 244 mm). Adapted from [3]. 15

Figure 8: Variation of dimensionless critical velocity against dimensionless heat release rate. (O) measurements of critical velocity; (continuous line) equations (8) and (9): (dashed line) Thomas correlation (4). (from [29]). 17

Figure 9: Two step approximation of fire growth rate phase for the Second Benelux tunnel fire Tests and Runehamar Fire Test Program (from [18]) 22

Figure 10: A schematic of a hybrid computational grid for multiscale calculation of tunnel ventilation

flows and fires 25

Figure 11: Example of the network representation of a tunnel showing branches between nodes 33

Figure 12: Schematic of the control volumes adopted for the numerical solution 37

Figure 13: Meteorological pressure difference measured between the portals of Mont Blanc Tunnel [73] 45

Figure 14: Frejus tunnel: top) cross section; down) Schematic of the ventilation system layout 48

X

Figure 15: Schematic of the network used for the 1D calculation corresponding to the tunnel region between section T2 and T3 of Figure 14. 49

Figure 16: Velocity distribution computed with the developed 1D model and comparison to experimental data recorded in the Frejus tunnel. Longitudinal velocity as function of the tunnel length 50

Figure 17: Velocity distribution computed with the developed 1D model and comparison to experimental data recorded in the Frejus tunnel. Longitudinal velocity as function of the tunnel length at 4 different times 51

Figure 18: Schematic of the simplified fire representations used in this work. 80

Figure 19: Schematic of mesh used the fan representations used in this work. 83

Figure 20: Schematic of the CFD segregated solution algorithm. 84

Figure 21: Schematic of the Norfolk road tunnels cross section. 85

Figure 22: Schematic of the jet fan longitudinal position in the Westbound Norfolk road tunnel; jet fans are numbered from 13 to 24. 86

Figure 23: Examples of the different meshes used for half of the tunnel cross section and number of cells

per unit length of tunnel. 88

Figure 24: Comparison of the longitudinal velocity contours for meshes #1 to #4 in the tunnel at the

reference section 1. Velocity values are expressed in m/s. 88

Figure 25: Comparison of the longitudinal velocity contours for meshes #1 to #4 in the tunnel at the reference section 2. Velocity values are expressed in m/s. 89

Figure 26: Computed velocity profile in the tunnel for scenarios 1.1, 1.2, 1.3, 2.1, 2.2, 4.2, 5.1, 5.2, 6.1 from Table 8. The plotted velocity fields are relative to plane 1 of Figure 21. Velocity values are expressed in m/s. 91

Figure 27: Comparison between predicted velocity and experimental measurements provided by the

Sickflow 200 Units located at the centre of each tunnel tube. 92

XI

Figure 28: 3D visualisation of the computed velocity fields for ventilation scenario 3.4 involving all the 6 jet fan pairs. Velocity values are expressed in m/s. (not to scale). 93

Figure 29: Schematic of the experimental rig accordingly to Wu and Bakar [3]. Section B has been used in this study. 94

Figure 30: Examples of the different meshes used for half of the tunnel cross section and number of cells per unit length of tunnel. 95

Figure 31: Computed temperature and velocity fields for mesh #1 to #4 at reference sections 1 for a 30

kW fire at critical ventilation conditions. Temperature and velocity values are expressed in K and

m/s respectively. 96

Figure 32: Computed temperature and velocity fields for mesh #1 to #4 at reference sections 2 for a 30

kW fire at critical ventilation conditions. Temperature and velocity values are expressed in K and

m/s respectively. 97

Figure 33: Computed temperature and velocity fields in the vicinity of the fire source for a 30 kW fire at critical ventilation conditions. Temperature and velocity values are expressed in K and m/s

respectively. 98

Figure 34: Computed temperature and velocity fields at reference sections 1 and 2 for a 30 kW fire at

critical ventilation conditions. Temperature and velocity values are expressed in K and m/s respectively. 99

Figure 35: Computed temperature and velocity fields in the vicinity of the fire source for a 3 kW fire at critical ventilation conditions. Temperature and velocity values are expressed in K and m/s

respectively. 100

Figure 36: Computed temperature and velocity fields at reference sections 1 and 2 for a 3 kW fire at critical ventilation conditions. Temperature and velocity values are expressed in K and m/s

respectively. 101

Figure 37: Effect o fire Froude number on the predicted critical for a 3 kW and a 30 kW fire 102

Figure 38: up) Example computed velocity field for a pair of operating jet fans (jet fan discharge velocity ~34 m/s; down) Example computed temperature field for a 30MW fire subject to supercritical

XII

ventilation conditions. The velocity and temperature values are expressed in m/s and K, respectively. 107

Figure 39: Example of domain decomposition with and without overlapping [65]. 108

Figure 40: Example of domain decomposition for solution of Navier-Stokes problem using a Dirichlet-

Neumann iterative method. 110

Figure 41: Example of domain decomposition for solution of Navier-Stokes problem using a Schwartz

(Dirichlet-Dirichlet) iterative method. 111

Figure 42: Example of a domain decomposition in 1D and 3D sub-domains. 113

Figure 43: Visualization of a three stage coupling procedure. 117

Figure 44: Visualization of the interaction procedure between 1D and 3D grids at the left CFD domain

boundary (1D-CFD interfaces highlighted in green) 117

Figure 45: left) Evolution of total pressure and mass flow rate at a 1D-3D interface during a multiscale calculation. The maximum deviation allowed was 10-6. right). Deviation of the mass flow rate and total pressure at a 1D-CFD interface during a multiscale calculation 118

Figure 46: Diagram of the East and West Dartford Tunnels showing the relative positions of jet fans and extract shafts. (Drawn approximately to scale but with vertical distances five times larger) 122

Figure 47: East Dartford Tunnel; Picture taken approximately 1100 m from the Kent portal facing south (refer to Figure 46). 123

Figure 48: West Dartford Tunnel; Picture taken approximately 500 m from the Kent portal facing south (refer to Figure 46). 123

Figure 49: Layout and general dimensions of the tunnel cross sections (west tunnel to the left; East tunnel to the right) including the points 1-9 where the air velocities where measured (dimensions are expressed in mm). 125

Figure 50: Schematic of multiscale coupling between mono-dimensional and CFD models for the multiscale calculation of the jet fan discharge cone (1D-CFD interfaces highlighted in green) 127

XIII




Figure 54: Convergence of the predicted mass flow rate as a function of the location of the interface 131

Figure 55: Comparison of horizontal velocities between predictions (lines) and experimental measurements (symbols) in the West Tunnel. The two profiles and the numbers refer to locations in the tunnel section described in Figure 49. 132

Figure 56: Comparison of horizontal velocities between predictions (lines) and experimental measurements (symbols) in the East Tunnel. The two profiles and the numbers refer to locations in the tunnel section described in Figure 4.b. 134

Figure 57: Typical flow pattern produced by a series of seven jet fan pairs operating in the West Tunnel (not to scale). Velocity isocontours from 2 m/s to 20 m/s in steps of 2 m/s; Velocity expressed in m/s. 136

Figure 58: Computational mesh for the CFD module around the jet fans in the West (right) and East (left) tunnels. (Note: the West tunnels jet fans are installed in niches on the ceiling, in the East tunnel they are not.) 137

Figure 59: CFD calculated jet fan thrust vs. tunnel average velocity for the Dartford tunnels. 137

Figure 60: Comparison between experimental data and model predictions. 139

Figure 61: Results for the West Tunnel, using the strategy for Zone C (Kent supply on, Essex extract on), varying the number of active jet fan pairs. (Note: Zone C extends from approximately 700 m into the tunnel to 1370 m). 141

Figure 62: Results for the East Tunnel, using the strategy for Zone C (Kent supply on, Essex extract on), varying the number of active jet fans. (Note: Zone C extends from approximately 700 m into the tunnel to about 1300 m) 142

XIV

Figure 63: Layout of the tunnel used as case study showing the relative positions of the fire, jet fans and portals (Not to scale). 146

Figure 64: Schematic of the multiscale model of a 1.2 km tunnel including portals, jet fans, and the CFD domain of the fire region. Contours of the temperature field show the fire plume. (Not to scale). The 1D-CFD interfaces have been highlighted in green 148


Figure 66: Comparison of the longitudinal velocity (left) and temperature (right) contours for meshes #1 to #4 in the tunnel at Reference Section 1 for a 30 MW fire. The velocity and temperature values are expressed in m/s and K respectively. 150

Figure 67: Comparison of the longitudinal velocity (left) and temperature (right) contours for meshes #1 to #4 in the tunnel Reference Section 2 for a 30 MW fire. Velocity and temperature values are

expressed in m/s and K respectively 150

Figure 68: Effect of the CFD domain length, LCFD, on the average longitudinal velocity and temperature at the outlet boundary of the CFD module. Units are in m/s and K respectively. Note that the

shortest module length is 20 m. 152

Figure 69: Effect of the CFD domain length LCFD on the error for the average longitudinal velocity and average temperature. Results for top) Reference Section 1; bottom) Reference Section 2. Error calculated using Eq. (76). 153

Figure 70: Effect of the CFD domain length LCFD on the horizontal velocity and temperature fields at

Reference Section 1 for a 30MW fire. The velocity and temperature values are expressed in m/s and K respectively 154

Figure 71: Effect of the CFD domain length LCFD on the horizontal velocity and temperature fields at

Reference Section 2 for a 30MW fire. Velocity and temperature values are expressed in m/s and K respectively 154

Figure 72: Comparison of results near the fire for the multiscale and the full CFD simulations for a fire of

30 MW and ventilation scenario 1. Velocity and temperature values are expressed in m/s and K

respectively. The longitudinal coordinates start at the upstream boundary of the corresponding CFD domain. 157

XV

Figure 73: Comparison of results near the fire for the multiscale and the full CFD simulations for a fire of 30 MW and ventilation scenario 2. Velocity and temperature values are expressed in m/s and K

respectively. The longitudinal coordinates start at the upstream boundary of the corresponding CFD

domain. 158



domain. 159



domain. 160

Figure 76: Characteristic curves of a tunnel region 50 m long where an activated jet fan pair (and a single jet fan) is located: Pressure drop between inlet and outlet vs. Mass flow rate across the inlet. (CFD calculated). 161

Figure 77: Characteristic curves of the tunnel region 400 m long where the fire is located: Pressure drop

between inlet and outlet vs. Mass flow rate across the inlet (CFD calculated) 162

Figure 78: Longitudinal velocity iso-contours, calculated using full CFD for 10 operating jet fans pairs. (Not to scale) 163

Figure 79: Predictions of average velocity for cold flow scenarios. Comparison and error between multiscale and full CFD results. 164

Figure 80: Layout of the tunnel used as case study showing the relative positions of the fire, jet fans and portals (Not to scale). 171

Figure 81: Fire growth curve, delay phase and detection times considered in the time dependent multiscale simulations. The fire growth curve is based on the work of Carvel (2008) [18]. 172

Figure 82: Schematic of the multiscale model of a 1.2 km tunnel including portals, jet fans, and the CFD domain of the fire region. Contours of the temperature field show the fire plume. (Not to scale). The 1D-CFD interfaces have been highlighted in green 173

XVI

Figure 83: Time dependent evolution of the mass flow rate through the tunnel for scenario 1, 2 and 3 (see Table 17). The time to detection is 2 min. Supercritical conditions (vair> 3m/s) are reached after 244 s, 190 s and 160 s for scenario 2 and 3 respectively 175

Figure 84: Multiscale results in the vicinity of the fire computed 2 min after the fire outbreak for scenario

1, 2 and 3 (see Table 17). The ventilation system is about to be started. Velocity and temperature values are expressed in m/s and K respectively. The longitudinal coordinates start at the upstream

boundary of the corresponding CFD domain. (not to scale) 176

Figure 85: Temperature profiles computed by the multiscale model 3 min after the fire outbreak for scenario 1, 2 and 3 (see Table 17). The ventilation system is operative since 1 min. Temperature values are expressed in K. (not to scale) 176

Figure 86: Longitudinal velocity profiles computed by the multiscale model 3 min after the fire outbreak for scenario 1, 2 and 3 (see Table 17). The ventilation system is operative since 1 min. Velocity values are expressed in m/s. (not to scale) 177

Figure 87: Temperature and velocity profiles 5 min (left column) and 10 min (right column) after the fire outbreak for scenario 1, 2 and 3 (see Table 17). Temperature and velocity values are expressed in K. and m/s, respectively (not to scale) 178

Figure 88: Dependence between number of operating jet fan pairs, TD, and time required to remove back-layering computed from the moment of the ventilation system activation. 181

Figure 89: Dependence between number of operating jet fan pairs, TD, and time required to remove back-layering computed from the fire outbreak 182

XVII

LIST OF TABLES

Table 1: Extension of tunnels in Europe 1

Table 2: Approximate energy contents of typical tunnel fire loads [11,12] 4

Table 3: Approximate max HRR for typical tunnel fires 4

Table 4: Approximate smoke production from tunnel fires [12] 6

Table 5: Maximum peak temperature recorded on full scale experimental tunnel fires [1,2,14,19,20]. 6

Table 6: Summary of the observed correlation between ventilation rate, delay phase length and fire growth rate (from [18]). 22

Table 7: Summary of the published CFD studies related to tunnel fires discussed in the literature review. 66

Table 8: Summary of ventilation scenarios explored during the experimental campaign conducted in the Westbound Norfolk road tunnel. Scenarios having the measurement unit located in the vicinity of

an operating jet fan have been highlighted in grey. 86

Table 9: Grid Independence Study for a scenario involving an operating jet fan pair in the Norfolk tunnels 87

Table 10: Grid Independence Study for a scenario involving a 30 kW fire scenario 96

Table 11: Grid Independence Study for a scenario involving an operating jet fan pair in the West tunnel 128

Table 12: Summary of ventilation flows in the tunnels resulting from various ventilation strategies. The operating ventilation devices in each scenario are indicated by ON. The predicted ventilation

velocities in the incident zones are highlighted in bold 140

Table 13: Summary of ventilation and fire scenarios analysed with the multiscale technique 147

Table 14: Grid independence study of the full CFD domain for a 30 MW fire and 3 operating jet fan pairs. 149

Table 15: Comparison between Full CFD and Multiscale predictions for the 7 scenarios investigated. The multiscale results are obtained with direct coupling. The table presents only bulk flow data. 155

XVIII

Table 16: Comparison between Full CFD, Multiscale, and 1D model predictions for the 7 scenarios investigated. The multiscale results are obtained with indirect coupling. The table presents only

bulk flow data. 164

Table 17: Summary of the ventilation scenarios considered in the time dependent analysis 173

Table 18: Summary of the ventilation scenarios considered and numerical findings 180

XIX

ACKNOLEDGEMENTS

I would like to thank my supervisors Prof. Vittorio Verda and Prof. Romano Borchiellini for their constant support, their precious advices and feedbacks. They are the original masterminds behind this work. A special Thank You to Vittorio for the patience he had dealing with me during the last four years and above all and for being, besides an enthusiastic advisor, a good friend.

I would also like to thanks my supervisors Dr. Guillermo Rein and Prof. Jose Luis Torero for welcoming me at University of Edinburgh half the way through this project and for hosting me for more than 2 years. I thank them for tens of helpful discussions, for their invaluable feedback and for their contagious optimism. I have been very fortunate to learn about fire from them.

I owe my gratitude to Dr. Ricky Carvel for his expert comments and for providing insightful discussions. His help has been invaluable for the completion of this work.

I would like to thank Le Crossing Company Ltd., Jacob (UK) Ltd. and the Highways Agency for allowing access to the Dartford tunnels, assistance with the on-site measurements and permitting the completion of this work and the publication of several papers. Special thanks to Stuart Lowe of Jacobs (UK) Ltd. for all his help with the on-site measurements.

I am grateful to Transurban Ltd (AU) for supplying useful experimental measurements adding significant quality to this thesis. Special thanks to Cameron Torpy of Transurban Ltd (AU) for being a kind and expert interlocutor.

I would like to show my gratitude to all my colleagues in Torino, Adriano, Flavio, Chiara e Giorgia for many fruitful discussions but also for their friendship. The have been always around whenever I needed a break. A special thanks goes to Adriano for our innumerable conversations on CFD and music related topics and to Giorgia for being always so helpful while I was away.

XX

I am also thankful to all the people I have meet in the last years at University of

Edinburgh. Thanks to Nicolas and Cristin for all the interesting talks, the laughs and the culinary arguments we had since the beginning. They are among the best flatmates I have ever had. Thanks to Hubert and Wolfram for being always around for a talk, a joke or a serious discussion. I am thankful to Pedro for our endless conversations and for sharing his experiences. I cannot forget Thomas and his family for several enjoyable dinners and for introducing me to the pleasures of the Deutsche Sprache. Thanks to Albert for all the interesting research related conversations and for our long discussions on the Italo-Corsican culture. Thanks to all my office mates Paolo, Adam, Cecilia, Rory, Angus, Joanne, Kate and Susan for never denying a smile or a chat.

I am thankful to all my old friends and flatmates in Torino, Sergio, Michele, Antonio, Pierluigi and Michele for making, whenever I am there, my life pleasant and never boring. Thanks to them for all those late dinners, pool and poker games, sport activities and for always keeping the doors of their flats open for me to stay.

I also have to thank my father, my mother and my sister for encouraging and supporting me to follow my own choices and for insisting to invest in my education. I am grateful to them for all the sacrifices and efforts they did and they still do for me. To them I dedicate this thesis.

Finally, I would like to express my deepest thanks to Daniela. Thanks not only for coming into my life, but for the special person she is, for her enthusiastic support to my choices. Without her care and encouragements, this work would not have been successfully completed.

THANK YOU ALL

XXI

ABSTRACT

Tunnels represent a key part of world transportation system with a role both in people and freight transport. Past events show that fire poses a severe threat to safety in tunnels. Indeed in the past decades over four hundred people worldwide have died as a result of fires in road, rail and metro tunnels. In Europe alone, fires in tunnels have brought vital parts of the road network to a standstill and have cost the European economy billions of euros. Disasters like Mont Blanc tunnel (Italy, 1999) and the more recent three Channel Tunnel fires (2008, 2006 and 1996) show that tunnel fire emergencies must be managed by a global safety system and strategies capable of integrating detection, ventilation, evacuation and fire fighting response, keeping as low as possible damage to occupants, rescue teams and structures. Within this safety strategy, the ventilation system plays a crucial role because it takes charge of maintaining tenable conditions to allow safe evacuation and rescue procedures as well as fire fighting. The response of the ventilation system during a fire is a complex problem. The resulting air flow within a tunnel is dependent on the combination of the fire-induced flows and the active ventilation devices (jet fans, axial fans), tunnel layout, atmospheric conditions at the portals and the presence of vehicles.

The calculation of tunnel ventilation flows and fires is more economical and time efficient when done using numerical models but physical accuracy is an issue. Different modelling approaches can be used depending on the accuracy required and the resources available. If details of the flow field are needed, 2D or 3D computational fluid dynamics (CFD) tools can be used providing details of the flow behaviour around walls, flames, ventilation devices and obstructions. The computational cost of CFD is very high, even for medium size tunnels (few hundreds meters). If the analysis requires only bulk flow velocities, 1D models can be adopted. Their low computational cost favours large number of parametric studies involving broad range ventilation scenarios, portal conditions and fire sizes/locations.

XXII

Another class of methods, called multiscale methods, adopts different levels of complexity in the numerical representation of the system. Regions of interest are

described using more detailed models (i.e. CFD models), while the rest of the system can be represented using a simpler approach (i.e. 1D models). Multiscale methods are characterized by low computational complexity compared to full CFD models but provide the same accuracy. The much lower computational cost is of great engineering value, especially for parametric and sensitivity studies required in the design or assessment of ventilation and fire safety systems. Multiscale techniques are used here for the first time to model tunnel ventilation flows and fires.

This thesis provides in Chapter 1 a general introduction on the fundamentals of tunnel ventilation flows and fires. Chapter 2 contains a description of 1D models, and a case study on the Frejus tunnel (IT) involving some comparisons to experimental data. Chapter 3 discusses CFD techniques with an extensive review of the literature in the last 30 years. The chapter provides also two model validations for cold ventilation flows in the Norfolk Tunnels (AU) and fire induced flows in a small scale tunnel. Chapter 4 introduces multiscale methods and addresses the typical 1D-CFD coupling strategies. Chapter 5 applies multiscale modelling for cold flow steady-state scenarios in the Dartford Tunnels (UK) where a further validation against experimental data has been introduced. Chapter 6 present the calculations from coupling fire and ventilation flows in realistic modern tunnel layout and investigates the accuracy of the multiscale predictions as compared to full CFD. Chapter 7 represents application of multiscale computing techniques to transient problems involving the dynamic response of the ventilation system.

The multiscale model has been demonstrated to be a valid technique for the simulation of complex tunnel ventilation systems both in steady-state and time-dependent problems. It is as accurate as full CFD models and it can be successfully adopted to conduct parametric and sensitivity studies in long tunnels, to design ventilation systems, to assess system redundancy and the performance under different hazards conditions. Time-dependent simulations allow determining the evolution of hazardous zones in the tunnel domain or to determine the correct timing for the activation of fixed fire fighting systems. Another significant advantage is that it allows for full coupling of the fire and

XXIII

the whole tunnel domain including the ventilation devices. This allows for an accurate assessment of the fire throttling effect that is shown here to be significant and for a prediction of the minimum number of jet fans needed to cope with a certain fire size. Furthermore, it is firmly believed that the multiscale methodology represents the only feasible tool to conduct accurate simulations in tunnels longer than few kilometres, when the limitation of the computational cost becomes too restrictive.

Multiscale Modelling of Tunnel Ventilation Flows and Fires FRANCESCO COLELLA

1

1.1. Introduction

Tunnels represent a key part of world transportation system playing a fundamental role both in people and freight transportation system, especially in developed countries. Around the world most major cities and metropolitan areas have metro systems accounting for hundreds of kilometres of underground tunnels and networked system. Similarly, in some mountainous regions, tunnels represent a vital part of the network transportation system. At present, the overall length for operational transportation

tunnels throughout the whole of Europe is larger than 15000 km [1]. An overview on the extension of the underground transportation systems in Europe is given in Table 1 including road and rail tunnels.

Italy Austria Switzerland Germany France UK Norway SpainRailways 1200 105 360 380 650 220 260 750Roads 1160 210 140 70 180 30 370 100Total 2360 315 500 450 830 250 630 850

Table 1: Extension of tunnels in Europe

The issue of tunnel fire safety has become more important in the last decades due to the social impact of disaster like Kings Cross underground station in 1987 (31 deaths),

1 Introduction


2

Baku Underground fire in 1995 (289 deaths), Gotthard Tunnel in 2001 (11 deaths), Tauern Tunnel in 1999 (12 deaths), Mont Blanc Tunnel in 1999 (39 deaths), Frejus Tunnel in 2005 (2 deaths) and Channel tunnel fires in 1996, 2006 and 2008.

According to French statistics [2] it appears that there are only one or two car fires (per km of tunnel length) every hundred million cars passing through the tunnel. Same order of magnitude can be expected for fire involving heavy good vehicles (HGVs). In this case, 8 fires per hundred millions of HGVs are expected, but only one will be enough serious to produce damage to the structure [3]. On the basis of such values, one can expect that the chance of an accidental tunnel fire can be negligible. However given the high number of tunnels in Europe, their high traffic density (several millions of vehicles per year) and their length (sometimes up to several tens kilometres), the probability of accidental fires become significant. For instance statistics indicates that, on average, one fire incident occurred practically every month within the Elb tunnel in Germany, from 1990 to 1999. And this is not an isolate case. Indeed in the past decade over four hundred people worldwide have died as a result of fires in road, rail and metro tunnels. In Europe alone, fires in tunnels have destroyed over a hundred vehicles, brought vital parts of the road network to a standstill - in some instances for years - and have cost the European economy billions of euros [4]. This serious problem has the potential to get worse it the future due to the drastic increase in the volume of dangerous goods transported and in the number of new operative tunnels.

1.2. Fundamentals of tunnel fires

This section is intended to provide a general overview of the fundamentals of tunnel fires. Fire behaviour in tunnel as well as in compartment is different from the behaviour in open space (free burning conditions). In particular, due to the confined enclosure, the heat feedback from the walls and hot gases enhances the fire burning rate. Furthermore, for very intense enclosure fires the oxygen supply can be reduced inducing a change in the combustion regime from fuel-controlled (also over-ventilated fires) to ventilation-controlled (under-ventilated fires). In the last case the combustion process generate a large amount of incomplete combustion products and toxic effluents.


3

Ingason identifies three main differences between compartment fires and tunnel fires [5]. The first is related to the maximum heat release rate (HRR) that can be attained. Typically, in small compartment fires the maximum HRR is controlled by the

ventilation factor that can be calculated as oo hA [m5/2] where Ao and h0 are the area

and the height of the opening, respectively. In the case of the tunnel fires, given the size of the tunnel cross section and the air flow eventually delivered by the ventilation system, the oxygen supply to the fire zone is at least one order of magnitude larger than typical compartment fires. Therefore, in tunnel fire scenarios, the limiting factor to the

maximum HRR is not represented by the ventilation conditions but by the fuel available. Under-ventilated conditions can be only achieved in severe tunnel fires with multiple vehicles involved in the burning process.

The second difference is related to the likelihood of attaining flashover. Flashover is defined as a transition from a localized fire to the general conflagration within the compartment when all the fuel surfaces are burning [6], and limited by ventilation flows. External flames typically appear at the vents of the compartment. Indeed, flashover is unlikely to take place in a tunnel due to the large convective losses from the fire to the surroundings and lack of full containment of hot fire effluents. Nonetheless, it must be stressed that the ventilation system plays an important role in the development of a tunnel fire, especially during the under-ventilated regime [7].

The third difference is related to the smoke stratification. Early stage compartment fires are generally characterized by a buoyant layer of hot gases under the ceiling. The same smoke pattern can be observed in the early stages of tunnel fires but in absence of longitudinal ventilation. In this condition, the smoke front will spread away from the fire zone, cooling down and partially mixing with the air layer underneath. However, after a certain distance and time the smoke layer will descend and touch the road deck. The distance from the fire at which such phenomenon takes place is mainly dependent on the tunnel geometry and fire characteristics. The activation of the ventilation system generally produces important change in the structure of the smoke layer. Moderate ventilation velocities (< 3 m/s) generate a certain degree of back-layering in the fire upstream region while the stratification is lost in the fire downstream region. A more


4

detailed discussion on the interaction between ventilation system and smoke movements will be presented in the next sections.

Tunnel fires usually involve material from vehicles including seats, tyres, plastic material from the finishing, fuel from the tanks and eventually the loading. The latter can be very variable. Evaluations of the energy content for typical load involved in tunnel fires are presented in Table 2.

Type of vehicle Approx. energy content [MJ]Private car 3000-7000Public bus 41000

TIR fire load 65000HGV 88000 - 247000

Tanker with 50 m3 of petrol 1500000

Table 2: Approximate energy contents of typical tunnel fire loads [8,9]

Besides the global energy content other characteristics are required to assess the hazard of a given fire scenario. Typically the design of the ventilation system and structures requires an evaluation of the fire heat release rate (HRR), the smoke production and the temperature distribution and the maximum temperature at the tunnel walls.

Indeed the fire HRR represents the single most important variable to evaluate fire hazard [10] and its design value has a great influence on the tunnel construction and operating costs. Several guidelines have been formulated on the basis of large scale tests [8,11-13]. An overview is given in Table 3.

Type of vehicle Maximum HRR [MW]1 passenger car 2.5 - 5

2-3 passenger cars 81 van 151 bus 20

1 lorry with burning goods 20-301 HGV 70-200Tanker 200-300

Table 3: Approximate max HRR for typical tunnel fires

The time evolution of the fire HRR (i.e. growth rate) is another important parameter to be evaluated when designing a ventilation system or an evacuation procedure. This task is much more complicated and only rough estimations can be provided with the current state-of-the-art. Fire growth is indeed linked to flame spread. Flame spread is directly


5

dependent on material properties with geometry layout and ventilation conditions playing a crucial role. Material properties controlling flame spread can be evaluated by using small-scale flammability testing within an acceptable degree of accuracy for material ranking purposes [14]. However, the extrapolation of small scale data to predict full scale behaviour is also a critical point still under active research, especially when attempting to span multiple orders of length scales. Typical tunnel fires involve a wide range of material, including thermo-plastics which show complex melting and dripping behaviour with burning surfaces highly convoluted. In typical full-scale fire scenarios every burning face sees a variety of radiant fluxes coming from the fire plumes and from other hot surface. The resulting heat release rate of a full-scale object is the sum of the heat release rate from a complex distribution of melting and burning surface, seeing a full spectrum of heat fluxes [15]. In general this distribution depends on the particular geometric configuration and it is not unique.

The geometry of the fire load also is critical issue when evaluating flame spread and the consequent fire growth curve. In opposed spread the flame develops against the air flow. In this case the heated region of the material produced by the radiant feedback from the flame is small and then the flame propagates slowly and steadily. In the case of concurrent flame spread, the air flow and the flame spread direction are the same. In this scenario the heated region of the material produced by the flame has the same dimensions of the flame itself. Concurrent flame spread rate is in between one and two orders of magnitude larger than opposite spread rates [14] and it is self-accelerating. Tunnel fires experience a wide range of geometry and consequently different spread regimes are present at different stages.

Further complexity is added when introducing the effect of the ventilation system controlling the oxygen supply into the fuel bed, the flame shape and amount of heat which is re-irradiated back to the burning surfaces [16].

Given the large uncertainty incurring on flame spread from all the previous considerations, a meaningful prediction the fire growth is a complex task and only rough estimation can be provided with the current state of the art. Most of them are based on experimental evidences. For example observations of the above cited tunnel


6

fire experiments have shown that the typical t2 fire representation [6] does not explain the growth of any of the experimental data available, while a two-step linear approximation provided a better estimation [17]. During the first growth stage, the fire would grow slowly up to 12 MW, while during the second stage, the growth rate would be significantly higher (up to 15 MW/min). A more detailed explanation will be given in the following sections.

Same rough estimations can be provided for smoke production. Average values given by PIARC and confirmed by the EUREKA fire test program [12] are resumed in Table 4.

Type of vehicle PIARC EUREKA TESTpassenger car 20 -passenger van - 30

2 -3 passenger cars - -1 van - -

lorry without dangerous goods 60 50 -60HGV - -

Petrol tanker 100 - 200 -

Smoke flow [m3/s]

Table 4: Approximate smoke production from tunnel fires [9]

Temperature distributions and peak temperature attained during a tunnel fire scenario represent important variable for design purposes. Also in this case the actual knowledge is based on experimental data. Table 5 gives an overview of the maximum temperatures recorded during full scale tunnel fires including MTFTVP, EUREKA, Second Benelux tunnel test and Runehamar tunnel fire tests [11-13,18,19]. As it can be seen, the temperature ranges are quite large mainly depending on the specific conditions including ventilation conditions, fire load and tunnel cross section geometry. A larger set of experimental measurements of tunnel fire peak temperature is available in [19].

Type of vehicle Peak Temperature [C]passenger car 200-400

bus 700HGV 1000-1365

petrol tanker 1000 -1400

Table 5: Maximum peak temperature recorded on full scale experimental tunnel fires [11-13,18,19].

The test involving HGVs fires showed that the temperatures measured downstream of the fire were very high with flaming zone expanding up 70100 m. Such high


7

temperature could affect the entire tunnel ceiling downstream of the fire causing considerable spalling of the unprotected tunnel ceiling and eventually flame spread to other vehicles. Same considerations can be given for the upstream region.

1.3. The role of the ventilation system

The ventilation system plays a fundamental role in tunnel safety both in normal operating conditions and in case of fire. In normal operating conditions, the ventilation has to dilute contaminants emitted from the travelling vehicles keeping the air quality within safety levels for the tunnel users. The dilution of smoke will have a direct improvement on the tunnel visibility. The first attempts of installing mechanical ventilation systems in tunnels have been made in the 1920s. This was mainly triggered by the concern on the increasing temperature which was taking place in the underground metro system in New York and London [20]. Previously, the ventilation of such environments was accomplished by utilizing the piston effect produced by moving trains and it was enhanced by the presence of vertical shafts permitting a continuous exchange of air with the exterior. Analogously, the introduction of the first mechanical ventilation devices in road tunnels was triggered by the concern on air quality and the impact of exhaust gases emitted by internal combustion engines.

Due to the growing concern on tunnel fire safety, the ventilation system has gained great importance also in the management of emergency fire scenarios in tunnels. In

these cases it has the complex task of smoke management. Which ventilation system is

to be selected depends mainly of the tunnel layout and the fire safety strategies chosen for the specific tunnel. However, ventilation systems fall in two broad categories: natural and mechanical. In the first case, the air movement is induced by temperature or pressure gradients across the tunnel portals (i.e. due to meteorological effects) which have importance for long tunnels, and by the piston effect induced by the traffic itself. Mechanical ventilation systems instead, use complex combinations of fans, ducts and dampers for the scope. Depending on the configuration, mechanical ventilation systems are classified in longitudinal, fully transverse ventilation systems and semi-transverse ventilation systems. However for specific reasons (i.e. enhance smoke control capabilities) hybrid configurations can be encountered.


8

1.3.1. Natural ventilation systems

Natural ventilation systems manly rely on meteorological conditions and piston effect from moving vehicles to guarantee acceptable environment conditions within a tunnel. Meteorological conditions, including temperature and static pressure difference across tunnel portals as well as the effect of the wind, can have a significant impact in long tunnels. Eventually, natural ventilation phenomena can be promoted by including vertical shafts due to an enhanced chimney effect. Unfortunately none of the previous variables can be relied upon when designing tunnel ventilation strategies.

Same considerations can be drawn when considering the ventilation flows due to piston effects. Indeed, it depends on a large number of factors, vehicle speed, vehicle spacing, traffic direction, vehicle drag coefficient, and tunnel geometry, and as expected, many of them cannot be controlled. Small-scale experiments have demonstrated that the ratio between air bulk airflow velocity and vehicle velocity is mainly dependent on the traffic conditions and ranges between 15% and 26% [21]. Full scale measurements under various realistic traffic situations performed in a 1.8 km long tunnel in Taipei City provided lower values: the ratio between vehicles and bulk flow speed ranged between 2% and 7% when the traffic density varied between 2 and 20 vehicles per km of tunnel length and the average traffic velocity is 90 km/h [22]. Figure 1 depicts the typical correlation between traffic density and induced ventilation flows in a tunnel in the Taipei City tunnel.

Figure 1: Typical traffic flow and induced ventilation in the 1.8 tunnel in Taipei City. Traffic density and induced ventilation as presented in [22]

Similar values have been encountered for railway tunnels during the passage of a train [23]. However, the same authors confirmed that, in two-way traffic conditions, the


9

effectiveness of the piston effect is compromised and the ratio between bulk flow velocity and vehicle velocity is radically reduced.

For this reason, natural ventilation systems are applied to short tunnels. Depending on the specific national guidelines, the boundary between short and long tunnels ranges between 350 m 700 m in Germany or 400 m in UK [24].

In case of fire, smoke cannot be controlled due to the absence of mechanical ventilation devices, and naturally stratifies and spreads longitudinally along the tunnel. Due to stratification, the lower portion of the tunnel cross section is free of smoke promoting a

safe evacuation of the tunnel users. The depth of the smoke layer underneath the ceiling varies with fire size and fire growth rate, tunnel layout (i.e. dimensions, slope, and cross section), distance from the fire source and eventually with the natural ventilation phenomena (i.e. environment conditions and piston effects). Due to the heat losses through walls, mixing at the interface with the fresh air which is recirculated beneath the smoke layer, the natural smoke stratification breaks down after a certain distance and the vitiated gases occupy the entire tunnel cross section. The smoke recirculation towards the fire source induces also a serious deterioration of the environmental conditions in the vicinity of the fire. Experimental observations demonstrate that stable stratification can be maintained initially for a distance ranging between 400 m and 600 m from the fire [24]. Eventually, the presence of intermediate chimneys can improve the smoke removal from the tunnel but usually this is not a reliable approach. For this reason, it is easy to understand that natural ventilation becomes significantly risky for long tunnels and it represents a viable approach only for tunnels shorter than few hundred meters.

1.3.2. Mechanical ventilation systems

1.3.2.1. Longitudinal ventilation systems

Longitudinal ventilation systems are designed in order to generate a longitudinal ventilation flow within the tunnel with air introduced or extracted from a limited number of points. The longitudinal movement can be induced by the presence of air injection points into the tunnel or by using fans installed on the ceiling providing


10

longitudinal thrust. The first design option uses Saccardo nozzles located in the vicinity of the tunnel portals which inject air with high velocity and induce longitudinal ventilation flow. A schematic of a Saccardo longitudinal ventilation system is depicted in Figure 2.

Figure 2: A schematic of a Saccardo longitudinal ventilation system [25]

Longitudinal ventilation systems based on jet fans use series of axial fans (known as jet fans or boosters) installed on the tunnel ceiling characterized by high thrust (hundreds of N) and high discharge ventilation velocities (around 30 m/s). The jet fans can be installed individually, in pairs or even more. A schematic of a jet fan longitudinal ventilation system is depicted in Figure 3.

Figure 3: A schematic of a jet fan longitudinal ventilation system [25]

Both the previous ventilation systems are characterized by an almost uniform ventilation velocity through the whole tunnel domain with pollutant concentrations and air temperature increasing in direction of the ventilation flows. In comparison to other more complex ventilation systems (i.e. transverse and semi-transverse ventilation system), longitudinal ventilation systems require less space for ventilation building and ductworks, and a lower capital investment. On the contrary, the tunnel cross section has to be large enough to accommodate their installation. The maintenance and operating cost break-even point associated with a large number of jet fans must be considered. If the system is characterized by a number of jet fans larger than 20, it may be economically convenient to move to other centralized ventilation layouts [26].


11

The typical ventilation strategies adopted in longitudinally ventilated tunnel require the ventilation system to push the smoke downstream of the incident region in the same direction as the road traffic flow, avoiding the smoke spreading against the ventilation flow (back-layering effect). The vehicles downstream of the fire zone are assumed to leave the tunnel safely. All the studies on back layering show that the maximum critical velocity is in the range from 2.5 m/s to 3 m/s [27-30]. Thus, an adequate ventilation system must guarantee air velocities higher than this range in the region of the fire

incident. A more detailed overview on the critical velocity will be given in the following sections. Longitudinal ventilation systems are very effective for tunnel with uni-directional traffic flows, providing enhanced smoke control for a wide range of fire sizes. The ventilation strategies to be adopted are also straightforward. Nowadays, their applicability is limited mainly by the tunnel length.

1.3.2.2. Transverse ventilation systems

Transverse ventilation systems are characterized by uniform air supply and extraction along the tunnel length realized by means of full-length ducts. Supply ducts are usually located either beneath the road deck or above a false ceiling and are connected to the tunnel environment through grills or dampers that can be automatically opened in specific location to promote smoke extraction. The ducts lead to ventilation stations equipped with axial fans. A schematic of a transverse ventilation system is presented in Figure 4. In long tunnels the supply ducts are usually divided in sections in order to limit the size of each ventilation station and the air velocities. Given the dimensions of the duct work and the size of the ventilation stations, the initial investment cost is high.

In normal operating conditions the concentration of pollutants is uniform along the tunnel length (if there is no longitudinal air flow) making this systems well suited for long tunnels also for bi-directional traffic operation.

In case of fire, the ventilation system is operated in order to maintain a smoke clear zone for evacuation purposes by creating a stable stratification of the smoke. The latter is extracted through dampers which are opened in the vicinity of the fire. Eventually fresh air can be supplied. More complex ventilation strategies can be used depending on the specific tunnel layout or boundary conditions.


12

Globally, transverse ventilation systems are operated in order to avoid the smoke spreading in the tunnel by promoting smoke confinement, stratification and extraction. An optimum strategy would provide limited air velocity (~ 1 m/s) in the fire vicinity. A velocity profile converging towards the fire zone is also desired in order to promote faster smoke confinement. Transverse ventilation systems are proved to be effective for smoke control in case of relatively small fires (< 20 MW). In these scenarios, the extraction efficiency appears to depend mainly on the air flow velocity while the shape of the dampers, for equal opening area, does not have any significant effect [31]. The same authors show that the efficiency of transverse ventilation systems mainly depends on the air flow velocity for small fire size. However, ineffective smoke and temperature managements have been observed for larger fire sizes [11].

It is worth to note that a viable longitudinal flow control is difficult to achieve, even if the system has a large capacity because there are not compensating forces acting in the longitudinal direction. Fire detection and localization are also critical issues for transverse ventilation system.

Figure 4: A schematic of a fully transverse ventilation system

1.3.2.3. Semi-transverse ventilation systems

Transverse ventilation systems are characterized by uniform air supply or extraction along the tunnel length realized by means of one full-length duct. Depending on the way the ventilation system is operated, semi-transverse ventilation systems can be classified as supply semi-transverse ventilation systems (see Figure 5) or exhaust semi-transverse ventilation systems (see Figure 6). The former are characterized by a uniform air supply while the latter have a uniform collection of air along the tunnel length. In normal

operating conditions supply semi-transverse systems are activated in order to provide


13

dilution to the traffic pollution. In emergency conditions the air supply could be used to dilute fire effluents; however, reversible fans should be preferably adopted and used to extract smoke during fire scenarios. In fire scenarios, exhaust semi-transverse systems are operated to extract smoke promoting smoke stratification and extraction.

The same limitations presented for fully transverse ventilation systems apply to semi-transverse systems. They have limited capability in controlling longitudinal ventilation flows and they are likely to be unable in managing smoke and temperature in large fire scenarios.

Figure 5: A schematic of a supply semi-transverse ventilation system

Figure 6: A schematic of a exhaust semi-transverse ventilation system

1.3.3. Hybrid ventilation systems

Beside the previous classification, ventilation systems with intermediate characteristics are often encountered worldwide. In most of the cases they are hybrid combinations of longitudinal and transversal layouts resulting from refurbishments or updating of old un-effective ventilations systems. This is the case of the Mont Blanc tunnel (11.6 km), which has been converted, after the catastrophic fire in 1999, from fully transverse ventilation system to hybrid transverse-longitudinal. Another example is represented by the Dartford Tunnels (UK) converted from semi-transverse ventilation system to hybrid semi-transverse-longitudinal. In both the previous cases the existing ventilation systems


14

have been updated with the introduction of jet fans for enhancing longitudinal smoke control.

In general hybrid ventilation systems are designed in order to provide high smoke control capabilities both in bi-directional and uni-directional traffic operation. In some cases they are operated in order to generate smoke-clear zones on both sides of fire site. The ventilation strategies used in hybrid ventilation systems are generally very complex requiring a careful analysis of all the variables involved including fire location, tunnel layout, boundary conditions at the portals and ventilation system settings.

1.4. Interaction between fire and ventilation system

The management of indoor ambient quality in underground structures both in ordinary operating and emergency conditions involves the use of the ventilation system.

Here it is stressed that a tunnel and the corresponding ventilation plant constitutes a single system. Its thermo-fluid-dynamic behaviour is affected by several internal and external factors, such as barometric pressure at the portals, tunnel slope, set-points of the ventilation system and traffic conditions [32]. Besides these, in emergency scenarios, fire dynamics, smoke movements, stratification and dilution, heat transfer with the tunnel linings are deeply coupled with the ventilation flows.

Mainly two aspects must be taken into account when considering the interaction between ventilation flows and fires: firstly, it controls the movements of smoke, stratification and dilution and secondly it supplies the fire with the oxidizer. A good understanding of the interaction between ventilation and a fire is therefore vital when developing a fire safety strategy.

1.4.1. Ventilation velocity and back-layering

The critical velocity is by definition the minimum longitudinal air flow required to prevent the occurrence of back-layering in tunnel fire scenarios. The back-layering phenomenon is the reverse smoke flow that can spread against the tunnel longitudinal ventilation if it is too low. An example of back-layering occurrence is depicted in Figure 7.


15

Ventilation flow

Figure 7: Photograph of a small scale tunnel fire during the occurrence of back-layering. The fire size in 15 kW. The tunnel has an arched cross section (width 274mm, height 244 mm). Adapted from

[33].

The exact value of the critical velocity depends mainly on the buoyant plume characteristics including smoke temperature, smoke flow rate, fire source size as well as tunnel height and width. The simplest techniques to predict the critical velocity are based on semi-empirical equations obtained by Froude Number preservation combined with some experimental data.

The Froude Number is defined as

gDU

forcesgravityforcesInertiaFr

2

== (1)

where g is the gravity, D and U are the characteristics length and velocity scales respectively. Equation (1) can be rearranged by using the density ratio of the smoke in order to include the effects of stratification. When rearranged in this for it is usually called Richardson number or modified Froude number:

=

=

FrvgDRi 12 (2)

where represents the density.

The first empirical relation based on Froude theory is due to Thomas (1958) [27] who argued that the characteristics of the flow are dependent on the ratio of buoyancy to inertial forces on the tunnel cross section. Thomas concluded that, when the ventilation


16

velocity is close to the critical value, the modified Froude number is close to 1 and therefore the back-layering does not occurs. Under this assumption, it can be written that

gHUc0

(3)

where, UC is the critical velocity value, H represents the tunnel height and 0 is the ambient temperature. After substituting an expression correlating the convective part of the fire heat release rate (HRR) and fire induced smoke characteristics (temperature, density and flow rate), a final correlation can be obtained

31

)1(

=

AcTHgQkU

pooc

(4)

where k is a proportionality constant, Q is the total HRR, To is the ambient temperature, cp is the air specific heat, A is the tunnel cross section and is the radiative fraction of the HRR. On the basis of experiment conducted in short corridors, the proportionality constant was found to be equal to 0.8 [34].

A similar correlation has been developed by Kennedy and co-workers:

31

)1(

=

ATcHgQKkU

pogc

(5)

AcUQTT

pcoof

)1( +=

(6)

( ) 8.0037.01 +=gk (7)

where K is an dimensionless empirical constant equal to 0.61, is the tunnel gradient and Tf is an average temperature of the fire effluents [35]. This correlation has been


17

built on the basis of small scale experiments conducted by Lee and co-workers in 1979 [36].

Thomas correlation is valid within a limited range of heat release rates where the 1/3 law well fits the experimental data. For higher heat release rates, the correlation fails because it is not able to represent the asymptotic behaviour of the critical velocity. Indeed, on the basis of small scale experiments, Oka and Atkinson pointed out that for high HRR the critical velocity reaches an asymptotic value which is independent from the HRR [28]. This behaviour is clearly presented in Figure 8 showing the correlation between dimensionless critical velocity and dimensionless heat release rate. Oka and Atkinson proposed a modified correlation whish is presented hereafter (equations from (8) to (11)):

Figure 8: Variation of dimensionless critical velocity against dimensionless heat release rate. (O) measurements of critical velocity; (continuous line) equations (8) and (9): (dashed line)

Thomas correlation (4). (from [28]).

31

*

*

12.0

=

QKU vc 12.0* Qfor (9)


18

where Q* and Uc* are the dimensionless heat release rate and dimensionless critical velocity that can be obtained by using equation (10) and equation (11). The proportionality constant Kv ranges between 0.22 and 0.38 depending on the burner geometry.

2/52/1*

HgcTQQ

poo= (10)

gHUU cc =

*

(11)

Such asymptotic behaviour has been also observed in full scale experimental campaigns such as Memorial tunnel fire ventilation test program (MTFVTP) [11] or EUREKA [12]. A theoretical explanation has been given by Wu and Bakar [33] attributing such behaviour to the positioning of the intermittent flames in the tunnel cross section. Indeed, free fire plumes are characterized by three different regimes [6]

1. persistent flame region, located close to the fire source and characterized by an accelerating flow of combustion gases

2. Intermittent flame region, characterized by intermittent flaming and a near-constant flow velocity

3. The buoyant plume characterized by a decreasing velocity and temperature with the height.

For relative small fires having flame length smaller that the tunnel height, only the

buoyant smoke impinges the ceiling and in under-ventilated conditions, it will generate back-layering. Obviously the characteristics of the buoyant plume will be depending on the fire HRR. However, for large enough fires, the intermittent flames will impinge the

ceiling occupying the upper portion of the tunnel cross section and in under-ventilated conditions they will be present in the back-layering. Intermittent flame are characterized by constant speed regardless the fire source and therefore, they build up a buoyancy force with is not sensitive to the fire HRR. Consequently the critical velocity will tend


19

to its asymptotic values. A similar explanation has been given also by Hwang and Edwards [37].

However, it must be stressed that simplified analysis based on Froude scaling theory cannot take into account the effect of the tunnel geometry (i.e. tunnel width) and tunnel slope on the critical velocity. Based on the classical Thomas theory it is easy to obtain a linear correlation between the critical velocity and the quantity (1/W)1/3 where W represents the tunnel width. Indeed, small scale experiments have confirmed that for aspect ratios greater than 1 (width W to height H) the critical velocity decreases with the tunnel width but following a trend different from the (1/W)1/3 law proposed by Thomas. Furthermore, it appears that for aspect ratios smaller than 1 the critical velocity

increases with the tunnel width [30]. Analogous deviations from the classical theory have been encountered when introducing blockages upstream the fire source or when varying the fire source geometry; in particular the critical velocity appears reducing when wider fire sources are adopted [28].

The effect of the tunnel slope on the critical velocity has been investigated by Atkinson and Wu [38] and by Ko and co-workers [39] on the basis of small scale experiments involving a propane gas burner for the former and methanol, acetone and n-heptane pool fires for the latter. In both the cases the results showed that the critical velocity increases with the tunnel slope due to the enhanced stack effect following equation (12)

( ) += KUU CC 10,, (12)

where ,CU and 0,CU are the critical velocities in a inclined and horizontal tunnel, the

tunnel slope, K an empirical constant ranging between 0.014 and 0.033 in accordance

with [38] and [39], respectively.

On the basis of the previous theoretical considerations supported by experimental measurements, it can be claimed that the maximum critical velocity value to be

expected in any tunnel fire scenario is between 2.5 m/s and 3 m/s. If the ventilation velocity is in this range (or eventually larger) the back-layering is usually avoided and


20

the smoke are pushed downstream of the fire region. Smoke stratification is usually compromised.

For ventilation velocities between 1 m/s and 2.53 m/s, depending on the fire source size, back-layering can occur. The back-layering distance usually varies between zero

and 17 times the tunnel hydraulic diameter [5]. For even lower ventilation velocities (between 0 m/s and 1 m/s) the back-layering distance can be very large (several hundred meters) and it is almost uniform in both directions.

Ingason proposed an approximated correlation to predict back-layering distance based

on small scale experiments [5] and Froude scaling theory.

3/1

3

HUTcgQ

HL

opo

b

(13)

Equation (13) correlates the back-layering distance Lb to the tunnel geometry, the ventilation velocity U and the fire HRR Q. The proportionality constant, deduced from small scale experiments, ranges between 0.6 and 2.2. Given the lack of large scale tests, great care must be adopted when predicting the back-layering distance on the basis of

equation (13). Indeed, in a recent work, it appears that equation (13) seriously under-predicts the back-layering distance (up to 1 order of magnitude) [40]. This conclusion has been drawn on the basis of a recent large scale set of experiments in a 1 km long tunnel (W ~ 10 m, H ~ 7 m, slope ~ 2%) involving fires between 1.8 MW and 3.2 MW.

1.4.1.1. Ventilation velocity and fire HRR

Ventilation flows have a direct impact on the tunnel fire dynamics. By using a probabilistic approach, Carvel and co-workers demonstrated that the HRR of a HGV

could increase in size by a factor 4 when the ventilation velocity is around 3 m/s and by a factor 10 when the ventilation is up to 10 m/s [41]. The authors found that a similar behaviour could be expected for the fire growth rate asserting that it can increase by a factor of 5 at 3 m/s and by a factor of 10 at 10 m/s.


21

Such behaviour is mainly dependent on the enhanced heat transfer from tilted flames and on the improved transport of oxygen into the fuel bed. However, it can be expected

that for ventilation flows higher that a certain limit, the cooling effect due to the ventilation flows counteracts against the improved radiative heat transfer from the

flames; in this conditions peak HRR and fire growth rate can be reduced.

The enhancing effects of the ventilation flows on the fire peak HRR and growth rate

have been observed experimentally both on large and small scale tests. In particular this behaviour has been recorded during the Second Benelux Tunnel fire tests for canvas

covered trucks loaded with wooden cribs and tyres. The fire growth rate with ventilation velocity ranging between 4 m/s 6 m/s was almost 2 times higher when compared to the fire development in no-ventilation scenarios. The peak HRR was about 1.5 times higher [13]. A similar behaviour has been observed on small scale experiments and described by Lonnemark and co-workers [16]. The increase in the peak HRR ranged between 1.31.7 and 1.82 times for high and low porosity wood cribs respectively.

They also found that the fire growth rate increased by a factor 5 to 10 depending on the tunnel cross section. Beyond a certain velocity limit the HRR and the fire growth rate

did not seem to vary significantly.

A more recent literature review presented by Carvel addressed other significant aspect

of the fire dynamics in tunnel [17]. The work reviewed a large number of tunnel fire experiments including the Second Benelux Tunnel fire tests [13], the Runehamar fire tests [8], and the EUREKA fire test program [12] and performed regular observations on the effect of the ventilation velocity on the fire growth phase.

The author observed that the typical t2 fire representation [6] was not fitting any of the experimental data and proposed a two-step linear approximation. During the first step

the fire would grow slowly up to 12 MW, while during the second step, the growth rate would be significantly higher (up to 15 MW/min). Figure 9 shows a two steps approximation of the fire growth phase as observed in [8] and [13].


22

Figure 9: Two step approximation of fire growth rate phase for the Second Benelux tunnel fire Tests and Runehamar Fire Test Program (from [17])

The changing in the fire regimes usually takes place after a delay phase usually as long as few minutes (from 2 to 6). The author observed also that the delay phase length and the fire growth rate are somehow correlated to the ventilation flows experienced by the

fire during its development. A table resuming the observed trends is introduced hereafter.

Table 6: Summary of the observed correlation between ventilation rate, delay phase length and fire growth rate (from [17]).

1.5. Analysis of tunnel ventilation systems and fires

On the basis of the previous discussions it is easy to understand that fire behaviour, smoke dynamics and ventilation flows are deeply coupled and they cannot be studied

separately. In other words, the resulting air flow within a tunnel is dependent on the combination of fire-induced flows, active ventilation devices (jet fans, axial fans), tunnel layout, atmospheric conditions at the portals and the presence of vehicles. Although an overall analysis of tunnel ventilation flows and fires can be very complex,

the resulting information is crucial for tunnel fire safety purposes. Studies of tunnel ventilation flows and fires are indeed fundamental to assess the capabilities of a


23

ventilation system to manage smoke, to design ventilation and evacuation strategies, to predict loss of tenability in the environment and to minimize damages to the structure.

Depending on the accuracy required and the resources available, a solution to the problem can be reached using different ways.

1.5.1. Small and large scale experiments

Full scale tests, generally conducted within unused tunnels, require very large financial investments but provide large amounts of collected data. Some examples have already

been cited. The Memorial Tunnel fire ventilation test program [11], the EUREKA fire test program [12] and the Second Benelux Tunnel fire test program [13] are only few examples. A wide review of the experimental tunnel fires conducted in the last 4 decades is available in [42]. Because of the huge costs associated, only a limited number of tests can be carried out. Furthermore they are highly specific and their outcome is strictly related to the specific tunnel layout, fire load material and geometry. Design

procedures sometimes use small scale tunnel models in order to represent ventilation and fire scenarios. Interpretation of their results is dependent on the relevant scaling

laws and model scale results may not have a general validity in relation to the full scale case. Nevertheless, experimental data are widely used to extrapolate proportionality constant used in semi-empirical correlation to predict back-layering occurrence and

distance, smoke production and smoke front velocity and temperature.

1.5.2. Numerical modelling

The analysis of tunnel ventilation systems can be also conducted using numerical

models based on a mathematical representation of the physical phenomena involved. Numerical models are usually highly flexible, significantly more economic than

experimental test, and allow for large parametrical studies and sensitivity analysis. The accuracy of numerical models must be always addressed on the basis of a direct

comparison of the results to experimental findings in order to assess range of applicability and limitations.

Several numerical approaches have been adopted by the international community to address tunnel fire safety issues.


24

The overall behaviour of the ventilation system can be approximated using 1D fluid dynamics models under the assumptions that all the fluid-dynamic quantities are

uniform in each tunnel cross section and gradients are only present in the longitudinal direction. 1D models have low computational requirements and are specially attractive

for parametric studies where a large number of simulations have to be conducted. In the last two decades several contributions on the application of 1D models to tunnel

ventilation flows and fires flows have been published; a literature review as well as a wide description of their accuracy and range of applicability will be presented in chapter

2.

Zone models are based on the experimental evidence that, under certain conditions, fire

effluents tend to stratify generating a cold air layer underneath and a hot smoke layer containing the fire effluents [43]. Zone models have been widely used to simulate compartment fires but their applicability in tunnel fire scenarios is limited. Indeed, they are not able to simulate tunnel smoke dynamics due to the lack of a dedicated horizontal

momentum equation needed to represent the longitudinal smoke transport in a tunnel environment. Furthermore, they are not able to take into account mixing between hot

and cold layers or to simulate fire scenarios where smoke stratification is lost (i.e. critical or supercritical ventilation scenarios).

Modified version of zone models have been developed trying to extend their use to tunnel fire scenarios. Charters and co-workers developed a modified version of zone

models having a three layer domain: a hot smoke layer, a mixing layer and a cold air layer underneath [44]. As for any zone model, the accuracy of the new one mainly relies on calibration constants needed to predict the mixing between layers, hot layer velocity and plume entrainment. A similar approach has been followed by Kunst who developed

a zone model and used it to predict back-layering [29]. Kunst model is in qualitative agreement with former, widely used models and it has been validated by comparison with mostly large-scale experiments in instrumented galleries. A more recent

application has been presented by Suzuki and co-workers [45]. The model uses several horizontal layers and provides reasonably accurate temperature distributions when

compared to small scale fire scenarios. However also in this case, the accuracy of the


25

model relies on calibration constants needed to predict plume entrainment and further validations test must be conducted.

CFD techniques are usually adopted in fire safety science when flow field data are needed. Such techniques are able to provide detailed temperature and velocity fields,

smoke movement and stratification, toxic species evolution, heat fluxes mapping, time to untenability conditions and other important variables. The computational cost of this

class of methods is high even for medium size tunnels and they are typically used for design verification. A literature review as well as a wide description of their accuracy

and range of applicability will be presented in chapter 3.

Another class of methods, called multiscale methods, adopt different levels of

complexity in the numerical representation of the system. The multi-scale concept is an

extension of the conventional 1D and CFD modelling techniques where the two models

are coupled together with the latter providing the boundary condition to the former and vice-versa. The multi-scale model is solved on a hybrid computational grid, where 1-

dimensional elements are linked to 3-dimensional ones generating a continuous domain in the streamwise direction (see Figure 10). The 3D elements are modelled by means of a CFD tool while 1D elements by using a conventional 1D model. During the solution procedure 1D and CFD models dynamically exchange information at the 1D-3D

interfaces and thus run in parallel. A literature review as well as a wide description multiscale modelling technique for tunnel ventilation flows and fires will be presented

in chapters from 4 to 7.

Figure 10: A schematic of a hybrid computational grid for multiscale calculation of tunnel ventilation flows and fires


26

1.6. Test cases

This thesis contains in different chapters several applications of 1D models, CFD

models and multiscale models. In most of the cases the developed models have been applied to predict the behaviour of real operative road tunnels. In some cases,

experimental campaigns have been undertaken to characterize the behaviour of tunnel and ventilation system. The collected data have been used to validate the developed

models and to assess their accuracy.

1.6.1. Case A: Frejus Tunnel, Bardonecchia (It) In Chapter 2, the Frejus tunnel behaviour is simulated with a 1D model. This tunnel is a two-way link between Italy and France with a total length of 12870 m and an

approximated hydraulic diameter of 6 m. The ventilation system is fully transverse and it is operated by means of full length supply and exhaust duct located over the tunnel

ceiling. Ordinary ventilation is operated by introducing fresh air along the tunnel through 3 U-shaped fresh air ducts which have 2 fans at each end. Fresh air openings

are installed each 5 m. Emergency ventilation is operated using the fresh air ducts and 3 U-shaped extraction ducts. The extraction dumpers are installed each about 130 m. A

more detailed description of the Frejus tunnel including typical emergency ventilation strategies will be given in chapter 2. Experimental data will be used to validate the

developed 1D model when simulating the tunnel ventilation system behaviour.

1.6.2. Case B: Norfolk road Tunnels, Sydney (Au) In Chapter 3 the Norfolk road tunnels ventilation systems are simulated by using a CFD

tool. These are two two-lanes unidirectional road tunnels located in Sydney (AU). The tunnels are 460 m long with a virtually flat gradient. Each tunnel, longitudinally ventilated, is equipped with 6 pairs of jet fans. A large set of air velocity measurements in the tunnel central section were made available by the tunnel operator and they have been used to validate the capability of CFD tools to model tunnel ventilation flows at

ambient conditions.


27

1.6.3. Case C: Wu-Bakar small scale tunnel

In Chapter 3 a CFD tool has been also used to simulate small-scale fire scenarios. Experimental data have been provided by Wu and Bakar [33] that carried out a series of small scale experiments on five horizontal tunnels with different cross-sections. They assessed, on the basis of accurate measurements in a controlled environment, the effect

on the critical velocity of tunnel cross section and fire heat release rate. Among the different cross sections, the data relative to the square cross-sectional tunnel (0.250.25 m

2 cross section) will be considered in this document. The small scale tunnel is around

15 m long and it is equipped with a circular porous bed propane burner (diameter equal to 0.106 m) located at a distance of 6.21 m from the tunnel inlet. The tunnel outlet is located at a distance of 8.7 m from the burner centre. The burner heat release rate,

controlled by the propane flow rate, was varied during the tests ranging between 1.5 kW and 30 kW. The measured values of critical velocities in two different fire scenarios (3 kW and 30 kW) will be used in the next sections to validate the fire CFD model.

1.6.4. Case D: Dartford Tunnels, London (UK) In chapter 5 the multiscale model has been used to simulate the ventilation flows in the Dartford tunnels. They are two twin-lane, uni-directional road tunnels under the River

Thames, crossing from Dartford at the south (Kent) side of the river to Thurrock at the north (Essex) side, about 15 miles east of London in the UK. Both tunnels have complex ventilation system consisting of a semi-transverse system together with

additional jet fans to control the longitudinal flow. Both the Dartford tunnels have two shafts with axial extraction fans located at relatively short distance from each of the tunnel portals. They length is around 1.5 km while the approximate internal diameter is 8.6 m and 9.5 for the West and the East tunnels, respectively. A more detailed d

network model tunnel ventilation

Documents

ventilation velocity

test case tunnel

case d

ventilation scenarios

velocity boundary conditions

case study

natural ventilation

mechanical ventilation